aarch64: Add back non-temporal load/stores from oryon-1's memset

I misunderstood the recommendation from the hardware team about non-temporal
load/stores. It is still recommended to use them in memset for large sizes. It
was not recommended for their use with device memory and memset is already
not valid to be used with device memory.

This reverts commit e6590f0c86632c36c9a784cf96075f4be2e920d2.
Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>
Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>

sysdeps/aarch64/multiarch/memset_oryon1.S

@@ -90,6 +90,8 @@ L(set_long):
 	cmp	count, 256
 	ccmp	valw, 0, 0, cs
 	b.eq	L(try_zva)
+	cmp	count, #32768
+	b.hi	L(set_long_with_nontemp)
 	/* Small-size or non-zero memset does not use DC ZVA. */
 	sub	count, dstend, dst
 
@@ -112,6 +114,30 @@ L(set_long):
 	stp	val, val, [dstend, -16]
 	ret
 
+L(set_long_with_nontemp):
+	/* Small-size or non-zero memset does not use DC ZVA. */
+	sub	count, dstend, dst
+
+	/* Adjust count and bias for loop. By subtracting extra 1 from count,
+	   it is easy to use tbz instruction to check whether loop tailing
+	   count is less than 33 bytes, so as to bypass 2 unnecessary stps. */
+	sub	count, count, 64+16+1
+
+1:	stnp	val, val, [dst, 16]
+	stnp	val, val, [dst, 32]
+	stnp	val, val, [dst, 48]
+	stnp	val, val, [dst, 64]
+	add	dst, dst, #64
+	subs	count, count, 64
+	b.hs	1b
+
+	tbz	count, 5, 1f	/* Remaining count is less than 33 bytes? */
+	stnp	val, val, [dst, 16]
+	stnp	val, val, [dst, 32]
+1:	stnp	val, val, [dstend, -32]
+	stnp	val, val, [dstend, -16]
+	ret
+
 L(try_zva):
 	/* Write the first and last 64 byte aligned block using stp rather
 	   than using DC ZVA as it is faster. */